A technique has been known in which a distributed file system for using the Hadoop and a plurality of virtual machines are constructed in a distribution system and distributed processing is performed to large-scale data. In the distributed file system, clustering is performed to a plurality of physical machines. PTL 1 discloses a technique for preventing that the replica data in a hadoop distributed file system (HDFS) is stored in one physical machine and a technique for making the replica data stored in the local physical machine preferentially read.